CRIMINAL RECOGNITION USING IMAGE RECOGNITION AND AI

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 06 | Jun 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 978
CRIMINAL RECOGNITION USING IMAGE RECOGNITION AND AI
Mrs. Shanmuga Priya.R., Harshavardhan.S.D, Monisha.H, Raushan Kumar Chaudhary,
H.K.Vinay Kumar
1Assistant Professor, Dept.of Computer Science, Dr.Ambedkar Institute of Technology, Karnataka, India
2Student, Dept.of Computer Science, Dr.Ambedkar Institute of Technology, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Criminal identification plays a vital role in
modern law enforcement, facilitating the detection,
apprehension and prosecution of persons involved in criminal
activity. Advances in criminal identification techniques have
been fueled by rapid advances in technology. Traditional
methods such as fingerprint analysis and DNA profiling have
been augmented by sophisticated automated systems that
enable faster and more accurate identification. Additionally,
advances in facial recognition, voice analysis, and gait
recognition have expanded the repertoire of tools available to
law enforcement agencies, improving their ability to identify
and apprehend criminals. This projectexploreshowtodevelop
a criminal identification system using ML and deep neural
networks to identify criminals at an airport. The faces of
criminals are stored in the database. Using Dlib68, people are
identified from the live camera feed. Detected persons are
compared with database images and identified as suspects.
After that, the picture of the wanted person with the date and
time will be sent to the authorized person by e-mail via the
SMTP protocol. The following method can be inherited as an
elegant way to ensure that enforcement is seamless.
Key Words: live camera feed, Dlib68, SMTP protocol
1.INTRODUCTION
The criminal record contains personal data about a specific
person together with a photo. In order to identify any
criminal, we need an eyewitness identification of that
person. Identification can be done using fingerprint, eyes,
DNA etc. One application is facial identification. The face is
the main focus of our attention in social interaction and
plays a major role in conveying identity and emotions. The
facial recognition system uses a database of stored images
and compares the captured image to find a match, if any. For
each face image, identification can be done usingRGBvalues
for eye color, face width, and height. This system is aimed at
identifying offenders in any investigativedepartment.Inthis
system, we store the criminal's image in the database.
Eyewitnesses select the slices that appear on thescreen,and
with this we get a face image from the database. Thus, this
system provides a very friendly environment for the
operator as well as the eyewitness to easily identify the
criminal if the crime record exists in the database. The
developed system is also the first milestone for image
detection and recognition based on surveillance video.
1.1 Proposed System
This project is focused on the development of an application
called Real-Time Criminal Identification System based on
facial recognition. We are able to detect and recognize
criminals from both image and video, a stream obtained
from images in real time. An authorized person logs in
through the welcome page to upload images for viewing.
Images are stored in a database. A web camera installed at
the airport records captured people and images. Image
processing and segmentation is performed on the captured
images. Using dlib's 68-point detection algorithm,onlyfaces
are identified and the identified images are compared with
the images stored in the database. If a match is found, the
captured image is sent to the authorized person along with
the date and time via email via SMTP. This saves a lot of time
and is a highly secure process and criminals can be easily
detected. Our app is 80 percent accurate and is fast, robust,
reliable and easy to use.
1.2 Objective
The main objective of real-time criminal identificationbased
on facial recognition application is to help police personnel
to identify criminals.
• The purpose of this application is to provide information
about a specific criminal that we are finding.
• Police personnel can use this appanytimeanywheretofind
the criminal.
• Any police personnel can access this application using
internet from anywhere anytime.
• We can also find criminals from live CCTV cameras.
• This application is fast, robust, reasonably simple and
accurate with a relatively simple and easier to understand
GUI

2. Architectural Diagram
Fig-1 : Architecture Design
2.1 System Module
There are six models used in this system. The Image
Preprocessing Module
 Image Segmentation Module
 Criminal Identification Module
 Database Module
 Image Matching Module
 Alert Module
1. Image preprocessing module
Processing features to be extracted to improve face
recognition speed. The face image is cropped and resized to
a smaller pixel value. If the images contain disturbances, it
will be difficult to train the model, resulting in an inaccurate
histogram
2. Image segmentation module
This is the main step. Various facial features are extracted.
The grayscale images from this step are used to identify the
criminal.
3. Criminal identification module
The extracted images are identified for faces using dlib
library and deep learning algorithm. So the person's face
alone is captured without any interruption of the
background.
4. Database module
It is a tool for collecting and organizing information.
Databases can store information about people, products,
orders, or anything else. Many pictures of criminals along
with their identities are stored in the database.
5. Image Matching Module
The resulting images are compared with existing images in
the database. If a match is found, data relatedtothatimage is
returned from the database, otherwise the recognized
person is not a criminal.
6. Warning module
If a criminal is detected, an alert message is sent to the
person along with the captured images and details of the
criminals using SMTP.
3. Working
The working theory for identifying criminals using CCTV
surveillance and image recognition revolves around using
the power of CCTV cameras and advancedimagerecognition
algorithms to identify and recognize criminals based on
captured images or video footage. The theory includes
several key components and steps, which are listed below:
1. CCTV surveillance system: Thebasisoftheworkingtheory
is a network of strategically placed CCTV cameras that
monitor public spaces such as streets, shopping centers or
transport hubs.Thesecamerascapturecontinuousvideoand
provide a huge amount of visual data for analysis.
2. Image capture: A CCTV system captures images or video
footage from recorded footage whenever there are
suspicious activities or events such as crimes or incidents
requiring investigation. These specific images serveasinput
for the subsequent recognition process.
3. Feature extraction: The next step involves extracting
characteristic visual features fromthepre-processedimages.
These features capture unique facial features, physical
attributes, or other identifying features that can help
distinguish individuals.
4. Recognition algorithm: The extracted features are then
used as inputs to the image recognition algorithm. The
algorithm compares the captured image with a database of
known criminals or a list of suspected persons. Various
recognition techniques such as similarity matching or
classification algorithms are used to identify potential
matches
5. Matching and Decision Making: The recognition algorithm
generates a similarity or probability score indicating the
likelihood of a match between the captured image and
database entries. A threshold is set to determine when a
match is considered significant enough to warrant further
investigation or intervention. Based on the similarity score,
the algorithm decides the presence of a potential criminal.

6. Alert and notification generation: If a significant match is
found, the system generates an alert or notification that is
sent to the relevant authorities or security personnel in real
time. An alert provides information about a potential
criminal, including their appearance, location and relevant
details, enabling quick action.
3.1 Facial Landmark Algorithm
The facial landmark algorithm is a computer vision
technique that aims to locate and identifyspecific landmarks
or facial points on the human face. These landmarks
represent key features such as the eyes, nose, mouth and
other facial structures.Byaccuratelyidentifyingandtracking
these landmarks, facial landmark algorithmsenablea variety
of applications, including facial analysis, facial recognition,
emotion detection, virtual reality avatars, and facial
animation. Facial landmark detection has 2 steps:
1. Detection of key facial structures on a person's face.
2. Includes face localization in the image.
We can perform face detection in many ways. We can
use OpenVMS built-in HearCascadeXML filesorevenTensor
Flow or using Keas. In particular, here we need to use HOG
(Histogram of Gradients) object detector and Linear SVM
(Support Vector Machines) specifically forthefacedetection
task. We can also do this using Deep Learning based
algorithms that are built forfacial localization.Thealgorithm
will also be used to detect faces in the image.
Fig-2: Face detection using facial landmark
3.2 D-Face Recognition Using LBP
Local Binary Pattern (LBP) It is a simple but very powerful
texture user that marks the pixels of an image with each
pixel of proximity and displays the resultasbinary numbers.
D-Face recognition involves extracting distinctive features
from face images using LBP. LBP captures local textural
patterns in facial regionsand providesrobustandexpressive
representations. In the context of criminal recognition, D-
Face recognition using LBP offers advantages such as
computational simplicity, robustness to illumination
changes, and insensitivity to facial expressions.ByusingLBP
in combination with a histogram, face images can be
described with a simple data vector.
Fig-3: PCA seeks directions that are efficient for
representing the data.
Local Binary Patterns, or LBPs for short, the work ofpopular
et al. such as texture descriptor, multi resolution grayscale
and rotation invariants
Texture classification with local binary patterns (LBP
definition was implemented in early 1993). Unlike Haralick,
it computes a global texture representation based on the
gray level concurrency matrix. LBPs compute the regional
representation of textures. A binary image is created by
comparing each pixel to its surrounding pixel. For example,
look at the original 3 x 3 pixel LBP neighborhood descriptor,
which works the same way:
Fig-4: Binary Pattern
The middle pixel and 8-bin numeric closure is the first step
to create a LBP for the neighborhood of 8 pixels.Intheimage
above, we select the center pixel and set it against its 8 pixel
surroundings. If the center of the pixel is largerthanthenext
line, set the value to l; set the value to 0, otherwise. A total
combination of 2 ^8= 256 possible LBP codes with eight
pixels will be achieved. With a 3x3 neighborhood, we have 8
neighbors that we need to check binary on. These binary
search results were stored in an 8-bit arrayandconverted to
decimal.

3.3 SMTP Protocol
SMTP protocol method
Store-and-Forward Method The store-and-forward method
is used within an organization.
End-to-End method The end-to-end method is mainly used
for communication between different organizations. The
SMTP client is the one that wants to send the mail and will
certainly contact the destination SMTP host directly in the
order to send the email to the destination.Thesessionisalso
initiated by the SMPT client. The SMTP server mainly
responds to the session request. So the session is started by
the SMTP-client and the SMTP-server responds to the
sender's request.
3.4 IMAGE PREPROCESSING
Image preprocessing is the steps taken to format images
before they are used in model training and inference. This
includes, but is not limited to, resizing, orientation,andcolor
corrections. Image preprocessing can also reduce model
training time and increase model inference speed. If the
input images are particularly large,reducingthesizeofthese
images dramatically reduces model training time without
significantly reducing model performance. For example, the
default image size on the iPhone 11 is 3024×4032. The
machine learning model Apple uses to create masks and use
portrait mode runs on images half that size before scalingits
output back to full size.
4. Applications
We can use this system anywhere for security purposes e.g.
In Law enforcement investigation, Public safety
5. Experimental Results
We have proposed a promising crime detection system for
live videos with faces. CCTV camerasareusedforcontinuous
video and image capture; on the main screen we will see
information about which image from the database
corresponds. When the image in the database matches the
image captured by the camera system, an email is senttothe
control room containing the details of the criminal and his
current location.
6. CONCLUSIONS
The real-time criminal identification system will help the
police to control the crime rate. This app helpstheminmany
different ways. With the advancementinsecuritytechnology
and the installation of cameras in publicareas,itwill become
easier for police personnel to track, trace and locate
criminals from the police control room using this app. In the
future, advanced facial recognitiontechniquescanbeusedto
improve the results and a login page must be created so that
any police personnel can remotely access this application.
Additionally, if a criminal is found in a certain zone, alert
messages should be sent to nearby police stations. The
developed application is simple and user-friendly. By using
advanced CSS styles and various front-end technologies, the
application interfacecanbedevelopedmoreaccordingto the
user's requirements.
REFERENCES
[1] Christopher, K., Port Security Management, Taylor &
Francis Group, FL, 2019.
[2] Järvinen, P., On Research Methods, Juvenes Print,
Tampere, 2019.
[3] Sertbay, H. and Toygar, Ö, “Face Recognition in the
Presence of Age Differences using Holistic and Subpattern-
based Approached”, in Proceedings of the 8th WSEAS
International Conference on Signal Processing (SIP '09),
Istanbul, Turkey, May 30 - June 1, 2019.
[4] Ramanathan, N., Chellappa, R. and Biswas, S.,
“Computational Methods for Modeling Facial Aging: A
Survey”, Journal ofVisual LanguagesandComputing,Volume
20, Issue 3, 2019.
[5] Saxholm,N.,Tilanneselvityslastiyksikköönkohdistuvasta
rikollisuudesta, BBA Thesis, Laurea University of Applied
Sciences, 2020. (In Finnish)
[6] Li, S. & Jain, A., Handbook of Face Recognition, Springer
Science + Business Media, Inc. USA, 2020.

[7] Belhumeur, P. N., Hespanha, J. P., & Kriegman, D. J.
Eigenfaces vs. Fisherfaces: Recognition Using Class Specific
Linear Projection.IEEETransactionsonPatternAnalysisand
Machine Intelligence. 19, pp. 711-720. IEEE Computer
Society, 2020.
[8] Bornet, O. (2019, May 19). Learning Based Computer
Vision with Intel's Open Source Computer Vision Library.
Retrieved April 2019.

CRIMINAL RECOGNITION USING IMAGE RECOGNITION AND AI

More Related Content

Similar to CRIMINAL RECOGNITION USING IMAGE RECOGNITION AND AI

More from IRJET Journal

Recently uploaded

CRIMINAL RECOGNITION USING IMAGE RECOGNITION AND AI